CN103823908B - Content recommendation method and server based on user preference - Google Patents

Content recommendation method and server based on user preference Download PDF

Info

Publication number
CN103823908B
CN103823908B CN201410108119.0A CN201410108119A CN103823908B CN 103823908 B CN103823908 B CN 103823908B CN 201410108119 A CN201410108119 A CN 201410108119A CN 103823908 B CN103823908 B CN 103823908B
Authority
CN
China
Prior art keywords
content item
recommendation
user
list
classification
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Expired - Fee Related
Application number
CN201410108119.0A
Other languages
Chinese (zh)
Other versions
CN103823908A (en
Inventor
王如章
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Beijing Feijiu Liutian Tech Co Ltd
Original Assignee
Beijing Feijiu Liutian Tech Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Beijing Feijiu Liutian Tech Co Ltd filed Critical Beijing Feijiu Liutian Tech Co Ltd
Priority to CN201410108119.0A priority Critical patent/CN103823908B/en
Publication of CN103823908A publication Critical patent/CN103823908A/en
Application granted granted Critical
Publication of CN103823908B publication Critical patent/CN103823908B/en
Expired - Fee Related legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0241—Advertisements
    • G06Q30/0251—Targeted advertisements
    • G06Q30/0269—Targeted advertisements based on user profile or attribute
    • G06Q30/0271—Personalized advertisement
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/95—Retrieval from the web
    • G06F16/953—Querying, e.g. by the use of web search engines
    • G06F16/9535—Search customisation based on user profiles and personalisation

Landscapes

  • Engineering & Computer Science (AREA)
  • Databases & Information Systems (AREA)
  • Theoretical Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Finance (AREA)
  • Strategic Management (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Accounting & Taxation (AREA)
  • Development Economics (AREA)
  • General Engineering & Computer Science (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Game Theory and Decision Science (AREA)
  • Data Mining & Analysis (AREA)
  • Economics (AREA)
  • Marketing (AREA)
  • General Business, Economics & Management (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

The invention provides a kind of content recommendation method and server based on user preference.This method includes：(a) the history preference information based on the user, calculates the preference function vector of the user；(b) based on preference function vector, the content item to be recommended is selected from recommendation list to be selected, to constitute consequently recommended list；And (c) provides the consequently recommended list to the user.

Description

Content recommendation method and server based on user preference
Technical field
The present invention relates to commending contents field, relate more specifically to the content recommendation method based on user preference and service Device.
Background technology
In recent years, with the popularization of the mobile terminals such as smart mobile phone, panel computer, mobile Internet have become we An indispensable part in production, life.Mobile terminal has no longer been the terminal of a basic communication and information transmission, But become the entertainment applications terminal that people carry with.This change, has expedited the emergence of huge Mobile solution market industry, people Various applications (for example, app) in daily study, work, rest on substantial amounts of use mobile terminal.For example, learning Various knowledge can be obtained during habit by educational app；Operationally it can obtain various important for example, by financial class app Financial Information, news etc. or pass through map class application and obtain positioning/navigation feature etc.；Media class can be passed through in rest App appreciates film, TV play, music, books etc..
At the same time, mobile terminal also changes the consumption pattern, consumption habit and consumer behavior of user.Pc user and shifting Dynamic terminal user has significant difference in time when they service in buying：Consumer on mobile terminal generally compares It is less patient, it is always desirable to can just to find the thing that they want at once, and pc user is then relatively more patient, is ready flower Time removal search waits more cheap but good commidities and/or service.
The exemplary of this respect is that the user of the 82% predetermined hotel room of utilization mobile terminal is within 24 hours Determine and complete.This is almost equivalent to user and has arrived destination just with mobile phone Lai Ding hotels.With predetermined hotel on computers The user in room compares, and these users want short many time spent.This " impulse buying " of mobile terminal user, " immediately Property purchase " behavior, be that one kind of conventional internet leisurely business model relatively is overturned in fact.For this new change Change, enterprise, which needs to help user to find them within the extremely short time, commodity interested and to be recommended, to capture shifting The first chance of dynamic marketing.Therefore it is accomplished by one kind and predicts what user may pay close attention to according to existing data (for example, sales data) The content recommendation system of commodity/service.
In existing commending contents scheme, collaborative filtering is one of important commending contents algorithm.GroupLens The collaborative filtering proposed in 1994 based on user (is referred to as CF-U, i.e. Collaborative Filtering User- Based) algorithm.The algorithm is the collaborative filtering being applied earliest, and it is broadly divided into three steps：
1) data are stated.User-project rating matrix R, the m row for generally obtaining a m × n first represents number of users, n row Represent item number, matrix element rI, jRepresent score values of the user i to project j；
2) similarity of user is calculated according to user-project rating matrix, is from big to small current use according to similarity Try to achieve an arest neighbors set N in family；And
3) recommending data collection is produced.Obtained for active user after k arest neighbors, can be based on this k neighbour to any The scoring of project, to predict scoring of the active user to the project.Then according to the height of prediction scoring, commented from all predict One or more projects are selected as last recommendation results in the project divided.
In this process, in order to obtain scoring most like k neighbours with active user, it is necessary to calculate the phase between user Like degree.Traditional method for measuring similarity is generally cosine similarity.User's scoring is treated as n-dimensional space vector.If user Project is not scored, then scoring of the user to the project is set to 0.Similarity between user a and b is scored by them Cosine angle between vector is measured to be given below：
Wherein, uaRepresent user a n dimension scoring vectors, ubRepresent user b n dimension scoring vectors, sim (ua, ub) represent ua And ubBetween similarity, cos (ua, ub) represent uaAnd ubBetween cosine similarity, | | represent vector norm, rA, jTable Show scorings of the user a to content item j.With traditional Forecasting Methodology, scorings of the prediction user a to project j, wherein N is user a's Arest neighbors set,It is user a average score,It is user b average score：
Wherein, pred (a, j ') represents scorings of the user a predicted to project j '.
It is increasingly huge however as the data volume in system, cause some current recommended technologies can not be effective real-time Make recommendation.At the same time, the problem of one annoyings commending system all the time is the sparse sex chromosome mosaicism of local data.Although system Conceptual data amount greatly, but divide to each user, its commodity for browsing and/or buying accounts for total commodity number in system Ratio is tangible very little, and which results in can not accomplish accurate and effective on the problem of calculating user's similarity.
For example, when personalized recommendation is done to user a, according to for example above-mentioned algorithm, user a personal behavior feature All embodied by his arest neighbors.In the case where there is adequate data, the interest that can embody to arest neighbors maximum probability user a is inclined It is good.But when data are excessively sparse, the user a found according to above-mentioned algorithm arest neighbors is real with user a similarity Very low on border, then the confidence level that user a behavioural characteristic is represented by arest neighbors is naturally also just had a greatly reduced quality.
In addition, above-mentioned algorithm also usually ignores user to this important information of the preference of variety classes commodity, so that also tight The validity of recommendation results is have impact on again.
At present, the application of comparative maturity includes Amazon and Netflix commending system in commending system field. Amazon commending systems are related to e-commerce field, have used mixing proposed algorithm, and one kind is improved project-based collaboration Filtering technique, another is the thing liked to user's commending friends on Amazon according to friend relation in user social contact network Product.Netflix commending systems are related to online movie rental, and equally a kind of combines of use improved after user behavior pattern Project-based collaborative filtering.The problem of both is respectively provided with the above to varying degrees.And relative to the two Field, in the field in Mobile solution market, because the concern that it is subject to is relatively fewer, therefore user behavior feature therein Achievement in research is also few, and Sparse Problem is more serious.
The content of the invention
In order to solve the above problems, there is provided the content recommendation method based on user preference according to the present invention and service Device.
There is provided a kind of content recommendation method based on user preference according to the first aspect of the invention.This method includes： (a) the history preference information based on the user, calculates the preference function vector of the user；(b) it is based on the preference function Vector, selects the content item to be recommended from recommendation list to be selected, to constitute consequently recommended list；And (c) is to the user The consequently recommended list is provided.
In certain embodiments, the history preference information is by counting in predetermined amount of time in user's download Hold item to obtain.
In certain embodiments, the predetermined amount of time is 1 month.
In certain embodiments, step (a) also includes：(a1) by the user counted download content item be divided into one or Multiple classifications；(a2) each classification in one or more of classifications is directed to, calculates in the content item that user downloads and belongs to this The total ratio for the content item that the number of the content item of classification is downloaded with user；And (a3) is according to one or more of classes The corresponding ratio of all categories in not, constitutes preference function vector of the user in the predetermined amount of time.
In certain embodiments, one or more of classifications include at least one of following：Audio, video, books, master Topic, game and tool software.
In certain embodiments, the recommendation list to be selected is to be generated by proposed algorithm for the user, and The proposed algorithm is at least one of following：Content-based recommendation；Recommendation based on collaborative filtering；Pushing away based on correlation rule Recommend；Recommendation based on effectiveness；Knowledge based engineering is recommended；And above-mentioned every any combination.
In certain embodiments, the recommendation based on collaborative filtering includes at least one of following：Collaboration based on user Filtered recommendation (CF-U)；Project-based collaborative filtering recommending (CF-I)；Collaborative filtering recommending (CF-M) based on model；And The every any combination of the above.
In certain embodiments, step (b) includes：(b1) according to the class of each content item in the recommendation list to be selected Not, the recommendation list to be selected is divided into one or more classification sublist；(b2) based on preference function vector, from corresponding Content item in the top is selected in classification sublist, to insert the consequently recommended list；And (b3) is based on described to be selected Non-selected content item in recommendation list, adjusts the consequently recommended list.
In certain embodiments, step (b2) includes：(b21) based on relative with each classification in preference function vector The preference function and the length of the consequently recommended list answered, are calculated in the consequently recommended list shared by each classification It is expected that number；And (b22) is directed to each classification, it is in the top from corresponding classification sublist to select corresponding expected numbers purpose Content item, to insert the consequently recommended list.
In certain embodiments, the selection in step (b22) is one of following two modes：Every time in a classification son row The content item of a current standings first is selected in table, and is carried out for all classification sublist circulations, wherein, skip and be chosen Empty classification sublist and the number of selected content item reach corresponding expected numbers purpose classification sublist；Or in each classification Corresponding expected numbers purpose content item is disposably selected in sublist, wherein, if the number of the content item in classification sublist is small In corresponding expected number, then all the elements in the classification sublist are all selected.
In certain embodiments, step (b3) includes：(b31) if the number of content item is less than in the classification sublist Corresponding expected number, content in the top is selected from the recommendation list to be selected in non-selected remaining content item , by the consequently recommended list completion；(b32) judge in the recommendation list to be selected in non-selected remaining content item Difference in the scoring of the content item ranked the first and the consequently recommended list between the scoring of the last content item is It is no to be more than predetermined threshold, if greater than the predetermined threshold, then exchange both and repeat step (b32) to be selected is pushed away until described Recommend the scoring of the content item ranked the first in list in non-selected remaining content item and ranking in the consequently recommended list Difference between the scoring of last content item is less than or equal to the predetermined threshold, to obtain the consequently recommended list.
In certain embodiments, the content item in the consequently recommended list be by content item marking order from high to low Arrangement.
In certain embodiments, the content item at least includes the following information：Content name, content type and Content is given a mark.
There is provided a kind of content recommendation service device based on user preference according to the second aspect of the invention.The service Device includes：Preference function computing unit, for the history preference information based on the user, calculates the preference function of the user Vector；Content recommendation select unit, for based on preference function vector, being selected from recommendation list to be selected in being recommended Rong Xiang, to constitute consequently recommended list；And recommendation list provides unit, for providing the consequently recommended row to the user Table.
In certain embodiments, the history preference information is by counting in predetermined amount of time in user's download Hold item to obtain.
In certain embodiments, the predetermined amount of time is 1 month.
In certain embodiments, the preference function computing unit is additionally operable to：(a1) by the user counted download Hold item and be divided into one or more classifications；(a2) each classification in one or more of classifications is directed to, calculates what user downloaded Belong to the number and the total ratio of the content item of user's download of the content item of the category in content item；And (a3) is according to institute The corresponding ratio of all categories in one or more classifications is stated, preference of the user in the predetermined amount of time is constituted and refers to Number vector.
In certain embodiments, one or more of classifications include at least one of following：Audio, video, books, master Topic, game and tool software.
In certain embodiments, the recommendation list to be selected is to be generated by proposed algorithm for the user, and The proposed algorithm is at least one of following：Content-based recommendation；Recommendation based on collaborative filtering；Pushing away based on correlation rule Recommend；Recommendation based on effectiveness；Knowledge based engineering is recommended；And above-mentioned every any combination.
In certain embodiments, the recommendation based on collaborative filtering includes at least one of following：Collaboration based on user Filtered recommendation (CF-U)；Project-based collaborative filtering recommending (CF-I)；Collaborative filtering recommending (CF-M) based on model；And The every any combination of the above.
In certain embodiments, the content recommendation select unit is additionally operable to：(b1) according in the recommendation list to be selected The classification of each content item, is divided into one or more classification sublist by the recommendation list to be selected；(b2) it is based on the preference Index vector, selects content item in the top, to insert the consequently recommended list from corresponding classification sublist；And (b3) non-selected content item in the recommendation list to be selected is based on, the consequently recommended list is adjusted.
In certain embodiments, the content recommendation select unit is additionally operable to：(b21) based in preference function vector The preference function corresponding with each classification and the length of the consequently recommended list, are calculated in the consequently recommended list Expection number shared by each classification；And (b22) is directed to each classification, corresponding expected numbers are selected from corresponding classification sublist Purpose content item in the top, to insert the consequently recommended list.
In certain embodiments, it is following two sides that the content recommendation select unit carries out selection from classification sublist One of formula：The content item of a current standings first is selected in a sublist of classifying every time, and for all classification row Table circulation is carried out, wherein, skip selected empty classification sublist and the number of selected content item reaches corresponding expected number Classification sublist；Or corresponding expected numbers purpose content item is disposably selected in each classification sublist, wherein, if point The number of content item in class sublist is less than corresponding expected number, then all selects all the elements in the classification sublist Go out.
In certain embodiments, the content recommendation select unit is additionally operable to：(b31) if in the classification sublist The number for holding item is less than corresponding expected number, the row of selection in non-selected remaining content item from the recommendation list to be selected The forward content item of name, by the consequently recommended list completion；(b32) judge non-selected in the recommendation list to be selected The scoring and the scoring of content item the last in the consequently recommended list of the content item ranked the first in remaining content item Between difference whether be more than predetermined threshold, if greater than the predetermined threshold, then exchange both and repeat step (b32) be straight The scoring of the content item ranked the first into the recommendation list to be selected in non-selected remaining content item is finally pushed away with described Recommend difference between the scoring of content item the last in list and be less than or equal to the predetermined threshold, to obtain described finally pushing away Recommend list.
In certain embodiments, the content item in the consequently recommended list be by content item marking order from high to low Arrangement.
In certain embodiments, the content item at least includes the following information：Content name, content type and Content is given a mark.
By using the method for the present invention and respective server, user interest preference can be introduced in commending system algorithm Information, enables last recommendation list to make personalized recommendation for the interest preference difference of different user.In addition, the present invention Target is that there is provided a kind of recommendation system for Mobile solution market on the basis of proposed algorithm that is existing or developing in the future The method of adjustment and server of many classification (generally higher than equal to 3) recommendation list of system.It can be existing or exploitation in the future The commending system algorithm service gone out, under the hardly premise of increase commending system added burden so that recommendation results are tried one's best Interest tendency with user is consistent with each other.
Specifically, the method according to the invention and server, with following some advantages：
(1) high divergence.Recommendation method/server of the present invention goes for any one recommendation system based on scoring System model.As long as the fraction on commodity (content item) can be provided, no matter then how particular technique is realized, recommendation side of the invention Method/server can use list adjustment algorithm to improve result.Therefore it is well positioned to meet demand of various businessmans, and System that is existing or purchasing in the future extensive adjustment need not be done to businessman；
(2) it is quick.In recommendation method/server of the present invention, user preference (example can be calculated using offline mode Such as, preference function vector), so as to improve the formation speed of Recommendations list, it is adaptable to which large-scale businessman is to recommending efficiency Demand；
(3) it is sensitive to user interest change reflection.The interest preference of client is in the relatively short period (for example, one Or some months) on it is stable, and can be changed on the relatively long period (for example, 1 year).The system is calculated using timing The mode of user interest preference, is elapsed over time, constantly adjustment user interest preference value, to meet the need that client constantly changes Ask；
(4) precision is improved.Be conducive to eliminating what Sparse was brought due to introducing the index vector of user interest preference Influence, so the recommendation list after adjustment will have bigger raising than the recommendation list before adjustment in accuracy.
Brief description of the drawings
By illustrating the preferred embodiments of the present invention below in conjunction with the accompanying drawings, above and other purpose, the spy of the present invention will be made Advantage of seeking peace is clearer, wherein：
Fig. 1 is the schematic diagram for showing the example application scenarios according to present disclosure commending system.
Fig. 2 is the example flow diagram for showing the commending contents based on user preference according to the present invention.
Fig. 3 is to show the example adjusted according to the recommendation list of the commending contents based on user preference of the present invention.
Fig. 4 is another the showing for showing the recommendation list adjustment according to the commending contents based on user preference of the present invention Example.
Fig. 5 is to show that the exemplary contents based on user preference according to embodiments of the present invention performed at server are pushed away Recommend the flow chart of method.
Fig. 6 is to show the block diagram for being used to perform the example server of the method shown in Fig. 5 according to embodiments of the present invention.
Embodiment
Before the specific design for introducing the present invention, by the first term simply introduced and be likely to occur herein, at this In text in the case of no clearly opposite instruction, these terms should be understood with implication described herein as.
Content：
In commending system field, the content of recommendation can be user's books interested, audio, video, theme, trip Play, tool software；Or can be any commodity and/or service etc. more generally, no matter its be charge or it is free, have It is shape or invisible, physical presence or virtual.
Mobile solution market：
All kinds of mobile phone application developers and its outstanding application are have accumulated, the real-time volume of different types of cellphone subscriber is met Test, download and order the comprehensive on-line mall of demand.It provides the user instrument by cell-phone customer terminal, wap and www websites etc. The one-stop services such as software, game, theme, video, audio, books.Famous Mobile solution market has：Google Play、 Apple Store etc..
Categorised content：
The classification carried out for content, total classification number be L (present application relates generally to situations more than 3 classes and 3 classes, when So, the application is applied equally to the situation of 1 class or 2 classes).For example：Will using the application for playing audio as its main propagation content It is classified as audio class and applies (for example, audio player)；The application of games is classified as game class application (for example, indignation is small Bird) etc..The method of classification can also be divided mathematically, the present invention is not limited to this with artificially defined.Change speech It, the present invention is applied to any number of classification and any mode classification.
Recommendation list：
, can beating based on other similar users content different to each when proposed algorithm is recommended for some user Point, the user is predicted to the marking of each different content.Select therein N number of interior in all contents that can be recommended Appearance is recommended to user together as final result, and such set is recommendation list.N is list length, and N is natural number. Recommendation order inside recommendation list is generally in no particular order, naturally it is also possible to be ranked up in some sequence, such as high by scoring It is low.
Redundancy recommendation list (or recommendation list to be selected)：
In the present invention, before length is produced for N consequently recommended list, various existing or exploitation in the future can be passed through The proposed algorithm gone out generates the redundancy recommendation list that a length is M (M >=N, and M is natural number), and consequently recommended list is by this Produced after the disclosed related algorithm adjustment redundancy recommendation list of invention.
Classification loop list：
Initial length is L (i.e. equal to total classification number), for record be described in detail below generating consequently recommended row The circular list optionally classified during filling is circulated during table.When a certain classification is unsatisfactory for fill conditions, by this Classification is removed from circular list.L is natural number.
Consequently recommended list：
List final lengths are N, and its initial content is sky, by the content item for meeting regularization condition in redundancy recommendation list Filled.After the completion of list adjustment, N number of content item of consequently recommended list is presented to user.
Collaborative filtering (CF, i.e. Collaborative Filtering)：
A kind of algorithms most in use in commending system field.The algorithm has two layers of meaning：Compared with the implication of narrow sense and broader Implication.In a broad sense, the collaboration that CF is used between multiple mechanisms, viewpoint, data source etc. was carried out to information or pattern Filter.The application of collaborative filtering is usually directed to very big data set.It is used to prospecting, the environmental monitoring of large area, finance data, Ecommerce and network application etc..
Narrowly say, CF is to predict that (filtering) is used by collecting preference or taste information (collaboration) from many users The method of the focus at family.The logic of its bottom is：If user a and user b has identical opinion in a certain content, For another content x, user a more likely has than another (having random opinion on content x) user c with user b Identical opinion.These predictions are specific for user, but actually they are according to from the anti-of a lot of other users What feedback was generated.Please note：This is not with being based simply on the recommendation that all users are made to the average opinion of a certain content With.
Specifically, the Algorithm Analysis user interest, finds several with the similarity degree highest of specified user in customer group Individual user's (arest neighbors), and comprehensive marking evaluation of these similar users to a certain content, user is specified to form system to this Marking (fancy grade) prediction to this content.Generally prediction fraction more Gao Ze represents user and more liked.According to this method, to institute There is the content that can be recommended all to be given a mark, and according to the sequence of prediction marking, one or more content is recommended to user.Beat The scoring information for evaluating collection is divided to may be not necessarily limited to of special interest, the record for content of especially loseing interest in also quite is weighed Will, so that for example the especially uninterested content of user is excluded outside recommendation list.
Collaborative filtering (CF-U, i.e. Collaborative Filtering User-based) based on user：
It is based on assuming as one " liking thing that similar people likes with you, you are likely to like ".So The main task of collaborative filtering based on user is exactly to find out the nearest-neighbors of user, so as to be made according to the hobby of nearest-neighbors The user is directed to the score in predicting of the unknown.
Project-based collaborative filtering (CF-I, i.e. Collaborative Filtering Item-based)：
By user to the scoring of different content project (item) come the similitude between evaluation item, based between project Similitude make recommendation.Collaborative filtering method based on project has a basic assumption " user can be caused emerging The project of interest, must be similar to the high project that score before it ", through the similitude between calculating project come instead of using person it Between similitude.
Interest preference：
In a period of time, mobile interchange network users can tend to use a certain class application.This and personal hobby, life Custom living has close contact.
Preference function vector：
Fancy grade of the specified user to a certain species content quantitatively is described with preference function, and by a certain user to institute There is L classes content (to be directed to a variety of classification, L is generally higher than the preference function vector that the preference function being equal to 3) constitutes the user.At this In application, by taking the application of six classes as an example, i.e. L=6, the preference function that user applies to game class is all applications downloaded In, game class application proportion.Similarly, the preference function applied to tool software, books, audio, theme, video class, i.e., For in all applications for being downloaded, respective classes application proportion.The inclined of user is constituted by the preference function of this six major class again Good index vector.In one example, preference function vector=(tool software, game, books, audio, theme, video).Than Such as, in 10 applications that user A is downloaded within a period of time (for example, one month), tool software class application has 3, and accounting is 0.3；Game class application has 5, and accounting is 0.5；Books class application has 0, and accounting is 0.0；Audio class application has 1, accounting For 0.1；Theme class application has 0, and accounting is 0.0；And video class application has 1, accounting is 0.1.The then preference of the user Index vector is (0.3,0.5,0.0,0.1,0.0,0.1).But in the present invention, content (application) classification it is in a unlimited number in 6, can also be less or more, i.e. the dimension of preference function vector is not limited to 6 dimensions or less or more.In addition, The order (that is, the order of content type) of each dimension is also not necessarily limited to (tool software, game, books, sound in preference function vector Frequently, theme, video) order, and can be any order, for example (theme, video, audio, tool software, game, books) Deng.
Tool software：
Refer to the other application in the application in addition to audio, video, books, game and theme class.For example, can wrap Include map class application, the application of financing class, address list management application, shopping class application etc..Sometimes also it is referred to as " software ".
With reference to the accompanying drawings to a preferred embodiment of the present invention will be described in detail, eliminate in the course of the description for this It is unnecessary details and function for invention, to prevent that the understanding of the present invention from causing to obscure.Hereinafter, it is applied to the present invention Exemplified by the scene of mobile radio system, to the present invention have been described in detail.But it is of the invention the invention is not limited in this Fixed communications, wired communication system can also be applied to, or applied to mobile radio system, fixed communication system Any mixed structure of system, wired communication system etc..For GSM, the invention is not limited in involved each The specific communication protocol of individual mobile communication terminal, can include but is not limited to 2G, 3G, 4G, 5G network, WCDMA, CDMA2000, TDSCDMA system etc., different mobile terminals can use identical communication protocol, it would however also be possible to employ different Communication protocol.In addition, the invention is not limited in the specific operating system of mobile terminal, can include but is not limited to iOS, Windows Phone, Symbian (Saipan), Android (Android) etc., different mobile terminals can be using identical operation System, it would however also be possible to employ different operating system., can be with addition, the invention is not limited in the specific operating system of server Windows, Unix, Linux, FreeBSD, Solaris of including but not limited to various versions etc., different server can be with Using identical operating system, it would however also be possible to employ different operating system.
Fig. 1 is the schematic diagram for showing the application scenarios according to present disclosure commending system 1000.As shown in figure 1, System 1000 can include terminal 100 and server 200.For the sake of clarity, illustrate only in figure a terminal 100, one Server 200, but the invention is not limited in this, terminal and/or server of two or more numbers etc. can be included.Eventually End 100 and server 200 can be communicated by communication network 300.The example of communication network 300 can include (but not limiting In)：Internet, mobile communications network, fixed circuit (such as xDSL, optical fiber) etc..
In the embodiment shown in fig. 1, in order to recommend on server 200 content, by the implementation according to the present invention Install on server 200 at the content recommendation service device end 250 (hereinafter referred to as server end 250) of example.Server end 250 can To be voluntarily arranged in the form of software in server 200 by service provider, or can be by production firm with hardware or solid The form of part is arranged in server 200.In certain embodiments, server end 250 can for example be purchased in service provider The application software dedicated for the present invention downloaded after server 200 from network is bought.In further embodiments, service Device end 250 can be the application program being for example pre-installed in by production firm with firmware or example, in hardware in server 200. In other embodiment, server end 250 can be the hardware module or server 200 itself produced by production firm.
In this embodiment, involved specific content recommendation is the various applications in Mobile solution market.This certain hair Bright not limited to this, i.e. content can also be other commodity and/or service, such as food, medicine, household electrical appliances, charges for water and electricity pay on behalf service, Fine pays on behalf service etc..Server 200 can recommend above-mentioned extensive stock and/or service to user.
In this embodiment, server 200 can be collected or otherwise obtain applying for a certain user and download first Historical information.A certain content (application) the classification quantity downloaded according to the user recorded in the historical information accounts for total download Percentage describes interest preference, and calculates preference function vector.At the same time or before this or afterwards, traditional association is passed through Recommended with filter algorithm for specific user, wherein, traditional collaborative filtering carries out user to each application and beaten Divide prediction, M application programs before marking prediction highest are selected, to constitute redundancy recommendation list (that is, recommendation list to be selected).It is logical Adjustment (for example, cyclic pac king, supplement replacement etc.) scheduling algorithm of the present invention is crossed, to generate consequently recommended list.And then cause most Types of applications ratio in whole recommendation list is tried one's best can be consistent with user preference index., will after adjustment operation all terminates Consequently recommended list recommends user as final result.
In one embodiment, if the user application program that (for example, 1 month) is downloaded within a period of time is 10 Individual, wherein tool software class application has 3, accounting 0.3；Game class application has 5, accounting 0.5；E-book class application has 0 It is individual, accounting 0.0；Audio class application has 1, accounting 0.1；Theme class application has 0, accounting 0.0；And video class application has 1 It is individual, accounting 0.1.Then the preference function vector of the user is (0.3,0.5,0.0,0.1,0.0,0.1).This index vector is also Reflect interest preference of the user in this month.The purpose of the present invention be desirable to finally into the recommendation list of user it is each The ratio of class application can show identical or at least close interest preference.It is equal to 10 in for example consequently recommended list length N In the case of, final recommendation list should be as far as possible by the 3 software class applications of marking highest, 5 game class applications, 0 figure Book application, 1 audio class application, 0 theme class application and 1 video class application composition.And traditional collaborative filtering base Such requirement is unable to reach on this, even if end product meets the ratio, coincidence is fallen within, and is not intended to so.So I Need to be adjusted recommendation list.
Next, reference Fig. 2 is specifically described the flow for commending contents by us.
In one embodiment, involved content recommendation is the various applications in Mobile solution market, then flow is substantially It is divided into following four basic execution steps：
Step one：Generate preference function vector.In one embodiment, historical record, meter are downloaded using the application of user Each user is calculated within nearest a period of time, the preference function vector applied for all categories.Each user can obtain One preference function vector of itself.
Step 2：Produce redundancy recommendation list.Entered using proposed algorithm that is existing or developing in the future on the user Row is recommended, and produces redundancy recommendation list, and the proposed algorithm that can be used is not limited to above-mentioned collaborative filtering, and it can be at least Use one of the following：Content-based recommendation, the recommendation based on collaborative filtering, the recommendation based on correlation rule, based on effect Recommendation, Knowledge based engineering are recommended and foregoing every any combination.If in addition, using collaborative filtering, its The collaborative filtering based on user is not limited to, it can also use the collaborative filtering recommending (CF-U) based on user, based on item Purpose collaborative filtering recommending (CF-I), the collaborative filtering recommending (CF-M) based on model and foregoing every any combination.Always It, when generating redundancy recommendation list, the present invention does not limit specifically used initial recommendation algorithm.
In addition, the content item in redundancy recommendation list generally includes the information of three aspects：It is content item (application) title, interior Hold the prediction marking of item (application) type and content item (application) in commending system.Redundancy recommendation list presses content type L classification sublist can be divided into.
Please note herein：The execution sequence of step one and step 2 not necessarily first carries out step one, is performing step Two, actually can also the two simultaneously (or substantially simultaneously) perform, or first carry out step 2 and perform step one again, the present invention is not It is limited to this.
Step 3：Generate consequently recommended list (list adjustment).According to the user preference index vector calculated in step one, Fill and be adjusted into consequently recommended list in the way of being described in detail below so that be of all categories in consequently recommended list The distribution of content is consistent to the preference of variety classes content with user or at least as far as possible consistent.
The method of foregoing adjustment can be divided into two steps：
(a) cyclic pac king, chooses Section 1 (score highest item) to final from redundancy recommendation list respectively classification successively Carry out inserting the filling of sequence method in recommendation list, and be constantly circulated between L classes.In one embodiment, the circulation can use one Individual classification loop list (length is L) is safeguarded.When in a certain classification sublist of redundancy recommendation list without project, or this point The number for the content item for having been filled with and (selecting) in class sublist reached the classification " (N is consequently recommended during preference function * N " The length of list), then the classification is removed from classification loop list, to avoid filling this point into consequently recommended list again The content item of class.When circular list is empty (that is, all classification has all been removed from circular list), cyclic pac king process knot Beam.
(b) supplement is replaced, after cyclic pac king terminates, by the whole projects of residue in redundancy recommendation list according to scoring by big To small sequence.When consequently recommended list less than when, the Section 1 in redundancy list is filled into by slotting sequence method successively consequently recommended Until filling up (completion) in list.After consequently recommended list is filled up, in computing redundancy recommendation list in current residual content item Scoring Section 1 scoring and the most last item of consequently recommended list it is (due to inserting the filling of sequence method, i.e., current in consequently recommended list Score lowest term) scoring between difference.The two is replaced when the difference is more than pre-set threshold value.Replacement method is, Consequently recommended list most last item is removed, and Section 1 in redundancy list is filled into consequently recommended list with slotting sequence method.It After compute repeatedly the difference and be replaced, until the difference is not more than threshold values, then adjusts process and complete, now consequently recommended List is consequently recommended list.
In another embodiment, it can also be filled without using " circulation ".For example, the sublist that can classify for each, From each classification sublist in disposably extract " the individual content items of preference function * N " insert consequently recommended list.If some is classified Content item number in sublist is less than " preference function * N ", then can equally will be finally in the stage follow-up " supplement is replaced " Recommendation list completion.
Step 4：Consequently recommended list is presented to user.For example, final by providing this from server 200 to terminal 100 Recommendation list, and provided by terminal 100 is shown to user.
In addition, in above-mentioned steps one, user can also be calculated according to other data related with content item to user Preference function vector.For example, range of information that can be according to produced by user in Mobile solution in the market activity, including under Carry, score, browsing, word comment etc. record, come calculate preference function vector.Certainly, user behavior data species at this moment is numerous It is many, and be not that each user profile is all our needs.It therefore, it can by behavioural characteristic extraction come garbled data. Finally, with reference to user attribute data, changed by behavioural characteristic, specific external data is converted into computer may be appreciated Behavioural characteristic vector.In view of user behavior real-time change, therefore, the part operation can generally be carried out in real time.
In addition, to the related external data of application can including applying recommendation tables, it include Apply Names, generic, The information of the sequence of user such as price concern.It is related that feature-content item is formed using recommendation tables bonding behavior characteristic vector Recommend.The a collection of application new due to having at regular intervals is added, therefore the part generally can regularly update.
In general, redundancy recommendation list can be produced by traditional proposed algorithm, and after recommendation list is adjusted Obtain last recommendation results.We need to explain content recommendation, and are presented to user together with recommendation results.
So far, with reference to the detailed commending contents flows according to embodiments of the present invention of Fig. 2.
Next, describing according to embodiments of the present invention for the final of user A and user B respectively with reference to Fig. 3 and Fig. 4 The example generating process of recommendation list.Fig. 3 and Fig. 4 all illustrate example redundancy recommendation list, the consequently recommended list of example and Example therein adjusts process.
User A：
Assume first that and be for user A redundancy recommendation list (Fig. 3 bottoms) length M generated by traditional collaborative filtering 20, and it is anticipated that the consequently recommended list length N of generation is 10 (Fig. 3 tops), total classification number L is 6, for example, count as described above The preference function vector of obtained user is (0.3,0.5,0.0,0.1,0.0,0.1), correspond to respectively (tool software, game, Books, audio, theme, video), list adjustment threshold value (scoring is poor) is 0.10.
(1) cyclic pac king：
Initialize classification loop list：[tool software, game, books, audio, theme, video].
Because the books class of user and the preference function of theme class are 0.0, therefore books class and theme class are followed from classification (0.0*10=0) is removed in circular row table, it is only remaining in classification loop list after removal：[tool software, game, audio, video]. Select tools software class Section 11, game class Section 16, audio class Section 1 14 and video class Section 1 18 successively Number, it is filled into slotting sequence method in consequently recommended list.
The number that audio class after filling has been filled with content item is 1, and it is more than or equal to 0.1 (audio-preferences index) * 10 (most Whole recommendation list length)=1.It is 1 that video class, which has been filled with item number, and it is more than or equal to 0.1*10=1.Therefore, by audio class and regard Frequency class is removed from classification loop list, i.e., only remaining：[software, game].Software class Section 1 No. 2 (now No. 1 is selected successively Not in redundancy recommendation list, other similarly), game class Section 17, be filled into slotting sequence method in consequently recommended list. Then, software class Section 13, game class Section 18 are selected successively, are filled into slotting sequence method in consequently recommended list.
Now, it is 3 that software class, which has been filled with item number, after filling, and it is more than or equal to 0.3*10=3.Software class is followed from classification Circular row table is removed, i.e., only remaining：[game].Game class Section 19 is selected, is filled into slotting sequence method in consequently recommended list. Game class Section 1 10 is selected, is filled into slotting sequence method in consequently recommended list.Now, game class has been filled with project after filling Number is more than or equal to 0.5*10=5 for 5.Game class is removed from classification loop list, i.e., it is only remaining：[].Now classification loop is arranged Table is empty, and cyclic pac king process terminates.
At this point it is possible to 10 content items in seeing the redundancy recommendation lists of Fig. 3 middle and lower parts to be indicated without diagonal line hatches It has been received in the consequently recommended list on top.
(2) supplement is replaced：
By remaining whole projects in redundancy recommendation list by descending sequence of scoring.Due to now consequently recommended list It is full, without carrying out supplement process.
Then, choose Section 1 16 in redundancy recommendation list in remaining content item (non-selected content item) with most It is 4.00-3.66=0.34 that the most last item 18 of whole recommendation list, which calculates scoring difference, and it is more than 0.10 (threshold values).Then this time shift Except most last item 18 in consequently recommended list, redundancy recommendation list Section 1 16 is inserted into consequently recommended list with slotting sequence method. Repeat said process, i.e. choose the current Section 1 of redundancy recommendation list 11 and the most last item 14 of consequently recommended list is calculated Difference is 3.99-3.79=0.20, and it is more than 0.10.Most last item 14 in consequently recommended list is then now removed, redundancy is recommended List Section 1 11 inserts consequently recommended list with slotting sequence method.Repeat said process, i.e. choose redundancy recommendation list current the One No. 17 are 3.88-3.81=0.07 with No. 10 calculating differences of most last item of consequently recommended list, and it is less than 0.10, therefore nothing It need to be replaced, replacement process terminates.
At this point it is possible to 2 content items indicated in seeing the redundancy recommendation lists of Fig. 3 middle and lower parts with diagonal line hatches by It is substituted into the consequently recommended list on top, and the diagonal line hatches in the middle of two lists are indicated in be replaced out two Rong Xiang.
After all adjustment processes terminate, consequently recommended list now seeks to the final result recommended to user.
User B：
Referring to Fig. 4, also assume that and the redundancy recommendation list of user B generations is directed to (under Fig. 4 by traditional collaborative filtering Portion) length M is 20, and it is anticipated that the consequently recommended list length N of generation is 10 (Fig. 4 tops), total classification number L is 6, for example The preference function vector for being computed as described above obtained user is (0.1,0.2,0.4,0.1,0.1,0.1), and (instrument is corresponded to respectively Software, game, books, audio, theme, video), list adjustment threshold value (scoring is poor) is 0.10.
(1) cyclic pac king：
Initialize classification loop list：[tool software, game, books, audio, theme, video].
Because the preference function of all categories is not 0.0, therefore select tools software class Section 11, game successively Class Section 16, books class Section 1 11, audio class Section 1 14, theme class Section 1 16 and video class Section 1 No. 18, it is filled into slotting sequence method in consequently recommended list.
Software class, audio class, the number for having been filled with content item of theme class and video class are 1, and it is more than or equal to 0.1 (preference function) * 10 (consequently recommended list length)=1.Therefore, software class, audio class, theme class and video class are followed from classification Removed in circular row table, i.e., it is only remaining：[game, books].(now No. 6 do not push away selection game class Section 17 in redundancy successively Recommend in list, other similarly), books class Section 1 12, be filled into slotting sequence method in consequently recommended list.
Now, it is 2 that game class, which has been filled with item number, after filling, and it is more than or equal to 0.2*10=2.Game class is followed from classification Circular row table is removed, i.e., only remaining：[books].Books class Section 1 13 is selected, is filled into slotting sequence method in consequently recommended list. Now, books class has been filled with item number for 3 less than 0.4*10=4 after filling.But because books class content item has all been selected into Consequently recommended list, therefore books class is removed from classification loop list, i.e., it is only remaining：[].Now classification loop list is Sky, cyclic pac king process terminates.
At this point it is possible in seeing the redundancy recommendation lists of Fig. 4 middle and lower parts with 9 content items being indicated without diagonal line hatches It is received in the consequently recommended list on top.
(2) supplement is replaced：
By remaining whole projects in redundancy recommendation list by descending sequence of scoring.Due to now consequently recommended list Less than, it is therefore desirable to carry out supplement process.During supplement, selection is scored most in remaining content item from redundancy recommendation list High content item, i.e. software class are currently scored highest item 2, and it is filled into consequently recommended list with slotting sequence method.Now, most Whole recommendation list is then supplemented process and terminated by completion.At this point it is possible to which seeing has the content item 2 of diagonal line hatches in Fig. 4 bottoms It is added in consequently recommended list.
Then, choose Section 18 in redundancy recommendation list in remaining content item (non-selected content item) with most It is 3.96-3.66=0.30 that the most last item 18 of whole recommendation list, which calculates scoring difference, and it is more than 0.10 (threshold values).Then this time shift Except most last item 18 in consequently recommended list, redundancy recommendation list Section 18 is inserted into consequently recommended list with slotting sequence method.Weight Multiple said process, i.e. choose the current Section 1 of redundancy recommendation list 3 and No. 13 calculating differences of most last item of consequently recommended list For 3.89-3.68=0.21, it is more than 0.10.Most last item 13 in consequently recommended list is then now removed, by redundancy recommendation list Section 13 inserts consequently recommended list with slotting sequence method.Repeat said process, i.e. choose the current Section 1 of redundancy recommendation list No. 17 are 3.88-3.79=0.09 with No. 14 calculating differences of most last item of consequently recommended list, and it is less than 0.10, therefore need not enter Row is replaced, and replacement process terminates.
At this point it is possible to 2 content items 3 indicated in seeing the redundancy recommendation lists of Fig. 4 middle and lower parts with diagonal line hatches In the consequently recommended list that top has been replaced to No. 8, and diagonal line hatches in the middle of two lists are indicated and are replaced out Two content items come.
After all adjustment processes terminate, consequently recommended list now seeks to the final result recommended to user.
So far, the specific example that commending contents are carried out for user A and user B is described in detail with reference to Fig. 3 and Fig. 4.
Fig. 5 is the flow for showing the content recommendation method 400 according to embodiments of the present invention performed in server 200 Figure.As shown in figure 5, method 400 can include step S410, S420 and S430.According to the present invention, some steps of method 400 Execution individually can be performed or combined, and can be performed parallel or order execution, it is not limited to the concrete operations shown in Fig. 5 Sequentially.In certain embodiments, method 400 can be as shown in Figure 1 server 200 or server end 250 thereon are performed.
Fig. 6 is the block diagram for showing example server 200 according to embodiments of the present invention.As shown in fig. 6, server 200 It can include：Preference function computing unit 210, content recommendation select unit 220 and recommendation list provide unit 230.
Preference function computing unit 210 can be used for the history preference information based on user, and the preference for calculating the user refers to Number vector.Preference function computing unit 210 can be the CPU (CPU) of server 200, digital signal processor (DSP), microprocessor, microcontroller etc., its can with the communications portion of server 200 (for example, radio receiving-transmitting unit, with Too network interface card, xDSL modems etc.) and/or storage part (for example, RAM, SD card etc.) be engaged, based on receiving and/or The history preference information of the user of storage, to calculate the preference function vector of the user.
Content recommendation select unit 220 can be used for based on preference function vector, and being selected from recommendation list to be selected to push away The content item recommended, to constitute consequently recommended list.Content recommendation select unit 220 can be that the center processing of server 200 is single First (CPU), digital signal processor (DSP), microprocessor, microcontroller etc., the communications portion that it can be with server 200 (for example, radio receiving-transmitting unit, Ethernet card, xDSL modems etc.) and/or storage part (for example, RAM, SD card etc.) phase Coordinate, based on preference function vector, the content item to be recommended is selected from the recommendation list to be selected for receiving and/or storing, with Constitute consequently recommended list.
Recommendation list provides unit 230 and can be used for providing a user consequently recommended list.Recommendation list provides unit 230 Can be the CPU (CPU) of server 200, digital signal processor (DSP), microprocessor, microcontroller etc., It can be engaged with the output par, c (for example, display, printer etc.) of server 200, provide a user by content recommendation The consequently recommended list that select unit 220 is generated.
Below with reference to Fig. 5 and Fig. 6, the content for being used to perform at server 200 according to embodiments of the present invention is pushed away Recommend method 400 and server 200 is described in detail.
Method 400 starts from step S410, in step S410, can by server 200 preference function computing unit The 210 history preference informations based on user, calculate the preference function vector of the user.
In the step s 420, preference function vector can be based on by the content recommendation select unit 220 of server 200, The content item to be recommended is selected from recommendation list to be selected, to constitute consequently recommended list.
In step S430, this can be provided a user by the recommendation list offer unit 230 of server 200 consequently recommended List.
In certain embodiments, history preference information can be by counting the content item that user downloads in predetermined amount of time Come what is obtained.
In certain embodiments, predetermined amount of time can be 1 month.
In certain embodiments, step S410 can also include being united by the preference function computing unit 210 of server 200 The content item that the user counted out downloads is divided into one or more classifications (S412)；For each class in one or more classifications Not, the number and the total ratio of the content item of user's download for the content item for belonging to the category in the content item that user downloads are calculated It is worth (S414)；And the corresponding ratio of all categories in one or more classifications, constitute user within a predetermined period of time Preference function it is vectorial (S416).
In certain embodiments, one or more classifications can include at least one of following：Audio, video, books, master Topic, game and tool software.
In certain embodiments, recommendation list to be selected can be generated by proposed algorithm for user, and recommend Algorithm can be at least one of following：Content-based recommendation；Recommendation based on collaborative filtering；Recommendation based on correlation rule； Recommendation based on effectiveness；Knowledge based engineering is recommended；And above-mentioned every any combination.
In certain embodiments, the recommendation based on collaborative filtering can include at least one of following：Collaboration based on user Filtered recommendation (CF-U)；Project-based collaborative filtering recommending (CF-I)；Collaborative filtering recommending (CF-M) based on model；And The every any combination of the above.
In certain embodiments, step S420 can include the content recommendation select unit 220 by server 200 according to treating The classification of the content item of each in recommendation list is selected, recommendation list to be selected is divided into one or more classification sublist (S422)；Base In preference function vector, content item in the top is selected from corresponding classification sublist, to insert consequently recommended list (S424)；And based on non-selected content item in recommendation list to be selected, adjust consequently recommended list (S426).
In certain embodiments, step S424 can include being based on partially by the content recommendation select unit 220 of server 200 The preference function corresponding with each classification and the length of consequently recommended list, are calculated in consequently recommended row in good index vector Expection number (S424-1) in table shared by each classification；And for each classification, phase is selected from corresponding classification sublist It is contemplated that the content item in the top of number, to insert consequently recommended list (S424-2).
In certain embodiments, the selection in step S424-2 can be one of following two modes：Every time at one point The content item of a current standings first is selected in class sublist, and is carried out for all classification sublist circulations, wherein, skip The classification sublist of sky has been chosen and the number of selected content item reaches corresponding expected numbers purpose classification sublist；Or every Disposably select corresponding expected numbers purpose content item in individual classification sublist, wherein, if the content item in classification sublist Number is less than corresponding expected number, then all selects all the elements in the classification sublist.
In certain embodiments, if the number of content item is less than corresponding expected number, step in classification sublist S426 can non-selected residue be interior from recommendation list to be selected including the content recommendation select unit 220 by server 200 Rong Xiangzhong selects content item in the top,
With by consequently recommended list completion：(S426-1)；Judge non-selected remaining content item in recommendation list to be selected In difference in the scoring of content item that ranks the first and consequently recommended list between the scoring of the last content item whether More than predetermined threshold, if greater than predetermined threshold, then exchange both and repeat step (S426-1) is until recommendation list to be selected In the scoring of content item that ranks the first in non-selected remaining content item and content the last in consequently recommended list Difference between the scoring of item is less than or equal to predetermined threshold, to obtain consequently recommended list (S426-2).
In certain embodiments, the content item in consequently recommended list can be by content item marking order from high to low Arrangement.
In certain embodiments, content item can at least include the following information：Content name, content type and Content is given a mark.
So far combined preferred embodiment invention has been described.It should be understood that those skilled in the art are not In the case of departing from the spirit and scope of the present invention, various other changes can be carried out, replaces and adds.Therefore, it is of the invention Scope be not limited to above-mentioned specific embodiment, and should be defined by the appended claims.

Claims (24)

1. a kind of content recommendation method based on user preference, including：
(a) the history preference information based on the user, calculates the preference function vector of the user, the preference function vector Indicate interest preference of the user to the ratio of types of applications；
(b) based on preference function vector, the content item to be recommended is selected from recommendation list to be selected, it is consequently recommended to constitute List；And
(c) the consequently recommended list is provided to the user,
Wherein, step (b) includes：
(b1) according to the classification of each content item in the recommendation list to be selected, the recommendation list to be selected is divided into one or many Individual classification sublist；
(b2) based on preference function vector, content item in the top is selected from corresponding classification sublist, to insert State consequently recommended list；And
(b3) non-selected content item in the recommendation list to be selected is based on, the consequently recommended list is adjusted.
2. according to the method described in claim 1, wherein, the history preference information is by counting in predetermined amount of time described Content item that user downloads is obtained.
3. method according to claim 2, wherein, the predetermined amount of time is 1 month.
4. method according to claim 3, wherein, step (a) also includes：
(a1) content item for downloading the user counted is divided into one or more classifications；
(a2) each classification in one or more of classifications is directed to, calculates in the content item that user downloads and belongs to the category The total ratio for the content item that the number of content item is downloaded with user；And
(a3) the corresponding ratio of all categories in one or more of classifications, constitutes the user in the pre- timing Between preference function vector in section.
5. method according to claim 4, wherein, one or more of classifications include at least one of following：Audio, regard Frequently, books, theme, game and tool software.
6. according to the method described in claim 1, wherein, the recommendation list to be selected be by proposed algorithm be directed to the user Generation, and the proposed algorithm is at least one of following：
Content-based recommendation；
Recommendation based on collaborative filtering；
Recommendation based on correlation rule；
Recommendation based on effectiveness；
Knowledge based engineering is recommended；And
Above-mentioned every any combination.
7. method according to claim 6, wherein, the recommendation based on collaborative filtering includes at least one of following：
Collaborative filtering recommending (CF-U) based on user；
Project-based collaborative filtering recommending (CF-I)；
Collaborative filtering recommending (CF-M) based on model；And
The every any combination of the above.
8. according to the method described in claim 1, wherein, step (b2) includes：
(b21) based on the preference function corresponding with each classification and the consequently recommended list in preference function vector Length, calculate expection number in the consequently recommended list shared by each classification；And
(b22) each classification is directed to, corresponding expected numbers purpose content item in the top is selected from corresponding classification sublist, with Insert the consequently recommended list.
9. method according to claim 8, wherein, the selection in step (b22) is one of following two modes：
The content item of a current standings first is selected in a sublist of classifying every time, and is followed for all classification sublist Ring is carried out, wherein, skip selected empty classification sublist and the number of selected content item reaches corresponding expected numbers purpose point Class sublist；Or
Corresponding expected numbers purpose content item is disposably selected in each classification sublist, wherein, if in classification sublist The number of content item is less than corresponding expected number, then all selects all the elements in the classification sublist.
10. method according to claim 8, wherein, step (b3) includes：
(b31) if the number of content item is less than corresponding expected number in the classification sublist, from the recommendation row to be selected Content item in the top is selected in non-selected remaining content item in table, by the consequently recommended list completion；
(b32) judge the scoring of content item that is ranked the first in the recommendation list to be selected in non-selected remaining content item with Whether the difference in the consequently recommended list between the scoring of the last content item is more than predetermined threshold, if greater than institute State predetermined threshold, then exchange both and repeat step (b32) until in the recommendation list to be selected it is non-selected it is remaining in In the scoring for the content item that Rong Xiangzhong ranks the first and the consequently recommended list between the scoring of the last content item Difference is less than or equal to the predetermined threshold, to obtain the consequently recommended list.
11. according to the method described in claim 1, wherein, the content item in the consequently recommended list is beating by content item Divide what order was arranged from high to low.
12. according to the method described in claim 1, wherein, the content item at least include the following information：Content name, Content type and content marking.
13. a kind of content recommendation service device based on user preference, including：
Preference function computing unit, for the history preference information based on the user, calculate the preference function of the user to Amount, the preference function vector indicates interest preference of the user to the ratio of types of applications；
Content recommendation select unit, for based on preference function vector, being selected from recommendation list to be selected in being recommended Rong Xiang, to constitute consequently recommended list；And
Recommendation list provides unit, for providing the consequently recommended list to the user,
Wherein, the content recommendation select unit is additionally operable to：
(b1) according to the classification of each content item in the recommendation list to be selected, the recommendation list to be selected is divided into one or many Individual classification sublist；
(b2) based on preference function vector, content item in the top is selected from corresponding classification sublist, to insert State consequently recommended list；And
(b3) non-selected content item in the recommendation list to be selected is based on, the consequently recommended list is adjusted.
14. server according to claim 13, wherein, the history preference information is by counting in predetermined amount of time Content item that the user downloads is obtained.
15. server according to claim 14, wherein, the predetermined amount of time is 1 month.
16. server according to claim 15, wherein, the preference function computing unit is additionally operable to：
(a1) content item for downloading the user counted is divided into one or more classifications；
(a2) each classification in one or more of classifications is directed to, calculates in the content item that user downloads and belongs to the category The total ratio for the content item that the number of content item is downloaded with user；And
(a3) the corresponding ratio of all categories in one or more of classifications, constitutes the user in the pre- timing Between preference function vector in section.
17. server according to claim 16, wherein, one or more of classifications include at least one of following：Sound Frequently, video, books, theme, game and tool software.
18. server according to claim 13, wherein, the recommendation list to be selected is for described by proposed algorithm User's generation, and the proposed algorithm is at least one of following：
Content-based recommendation；
Recommendation based on collaborative filtering；
Recommendation based on correlation rule；
Recommendation based on effectiveness；
Knowledge based engineering is recommended；And
Above-mentioned every any combination.
19. server according to claim 18, wherein, the recommendation based on collaborative filtering includes following at least one ：
Collaborative filtering recommending (CF-U) based on user；
Project-based collaborative filtering recommending (CF-I)；
Collaborative filtering recommending (CF-M) based on model；And
The every any combination of the above.
20. server according to claim 13, wherein, the content recommendation select unit is additionally operable to：
(b21) based on the preference function corresponding with each classification and the consequently recommended list in preference function vector Length, calculate expection number in the consequently recommended list shared by each classification；And
(b22) each classification is directed to, corresponding expected numbers purpose content item in the top is selected from corresponding classification sublist, with Insert the consequently recommended list.
21. server according to claim 20, wherein, the content recommendation select unit is carried out from classification sublist Selection is one of following two modes：
The content item of a current standings first is selected in a sublist of classifying every time, and is followed for all classification sublist Ring is carried out, wherein, skip selected empty classification sublist and the number of selected content item reaches corresponding expected numbers purpose point Class sublist；Or
Corresponding expected numbers purpose content item is disposably selected in each classification sublist, wherein, if in classification sublist The number of content item is less than corresponding expected number, then all selects all the elements in the classification sublist.
22. server according to claim 20, wherein, the content recommendation select unit is additionally operable to：
(b31) if the number of content item is less than corresponding expected number in the classification sublist, from the recommendation row to be selected Content item in the top is selected in non-selected remaining content item in table, by the consequently recommended list completion；
(b32) judge the scoring of content item that is ranked the first in the recommendation list to be selected in non-selected remaining content item with Whether the difference in the consequently recommended list between the scoring of the last content item is more than predetermined threshold, if greater than institute State predetermined threshold, then exchange both and repeat step (b32) until in the recommendation list to be selected it is non-selected it is remaining in In the scoring for the content item that Rong Xiangzhong ranks the first and the consequently recommended list between the scoring of the last content item Difference is less than or equal to the predetermined threshold, to obtain the consequently recommended list.
23. server according to claim 13, wherein, the content item in the consequently recommended list is by content item Marking order is arranged from high to low.
24. server according to claim 13, wherein, the content item at least includes the following information：Content name Claim, content type and content are given a mark.
CN201410108119.0A 2014-03-21 2014-03-21 Content recommendation method and server based on user preference Expired - Fee Related CN103823908B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201410108119.0A CN103823908B (en) 2014-03-21 2014-03-21 Content recommendation method and server based on user preference

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201410108119.0A CN103823908B (en) 2014-03-21 2014-03-21 Content recommendation method and server based on user preference

Publications (2)

Publication Number Publication Date
CN103823908A CN103823908A (en) 2014-05-28
CN103823908B true CN103823908B (en) 2017-07-14

Family

ID=50758972

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201410108119.0A Expired - Fee Related CN103823908B (en) 2014-03-21 2014-03-21 Content recommendation method and server based on user preference

Country Status (1)

Country Link
CN (1) CN103823908B (en)

Families Citing this family (40)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104168433B (en) * 2014-08-28 2018-02-09 广州华多网络科技有限公司 A kind of media content processing method and system
JP2016062239A (en) * 2014-09-17 2016-04-25 ソニー株式会社 Information processing device, information processing method and computer program
CN104301323B (en) * 2014-10-23 2017-10-31 中国科学院大学 Balanced third-party application personalized service and the method for user privacy information safety
CN105635210B (en) * 2014-10-30 2021-04-27 腾讯科技（武汉）有限公司 Network information recommendation method and device and reading system
CN104378659A (en) * 2014-10-31 2015-02-25 四川长虹电器股份有限公司 Personalization recommendation method based on smart television
CN105630474B (en) * 2014-11-06 2021-05-18 中兴通讯股份有限公司 Reminding method and device
CN104394439B (en) * 2014-11-20 2016-04-20 合一网络技术(北京)有限公司 Based on the system that the video recommendations of dynamic-configuration is shunted
CN104580385B (en) * 2014-12-16 2018-04-03 腾讯科技（深圳）有限公司 A kind of method and device for expanding customer relationship chain
CN104462560B (en) * 2014-12-25 2018-01-05 广东电子工业研究院有限公司 A kind of recommendation method of personalized recommendation system
CN105989056B (en) * 2015-02-06 2019-05-24 北京中搜云商网络技术有限公司 A kind of Chinese news recommender system
CN104683458B (en) * 2015-02-12 2018-07-06 广东欧珀移动通信有限公司 The method and server that a kind of wallpaper is recommended
CN106471491A (en) * 2015-05-29 2017-03-01 深圳市汇游智慧旅游网络有限公司 A kind of collaborative filtering recommending method of time-varying
CN104994431B (en) * 2015-07-06 2018-07-24 无锡天脉聚源传媒科技有限公司 A kind for the treatment of method and apparatus of web video attention rate
CN105404687A (en) * 2015-12-11 2016-03-16 重庆市科学技术研究院 Personalized recommendation method and system for learning behavior
US20170169500A1 (en) 2015-12-11 2017-06-15 Mastercard International Incorporated Systems and methods for generating recommendations using a corpus of data
WO2017124263A1 (en) * 2016-01-18 2017-07-27 杨瑛 Information pushing method when book is recommended and book recommendation system
CN107305559A (en) * 2016-04-21 2017-10-31 中国移动通信集团广东有限公司 Method and apparatus are recommended in one kind application
CN105975522B (en) * 2016-04-29 2019-06-21 清华大学深圳研究生院 A kind of multi-field content recommendation method and server
WO2018027737A1 (en) * 2016-08-11 2018-02-15 王志远 Method for pushing information according to change in commodity price, and pushing system
CN107730336A (en) * 2016-08-12 2018-02-23 苏宁云商集团股份有限公司 Commodity method for pushing and device in a kind of online transaction
CN106383907A (en) * 2016-09-30 2017-02-08 北京小米移动软件有限公司 Application recommendation method and device
CN106649733B (en) * 2016-12-23 2020-04-10 中山大学 Online video recommendation method based on wireless access point context classification and perception
TWI674544B (en) * 2016-12-29 2019-10-11 第一商業銀行股份有限公司 Intelligent product marketing method and system
CN107154109B (en) * 2017-06-02 2019-10-22 深圳正品创想科技有限公司 A kind of commodity rendering method, device and self-service cabinet
CN109088999B (en) * 2017-06-13 2021-04-06 中兴通讯股份有限公司 Method, device and computer readable storage medium for playing ring back tone
CN107506488A (en) * 2017-09-21 2017-12-22 程丹秋 A kind of intelligent music commending system
CN108108998A (en) * 2017-12-14 2018-06-01 百度在线网络技术（北京）有限公司 Showing advertisement method and apparatus, server, storage medium
CN108197219A (en) * 2017-12-28 2018-06-22 北京奇虎科技有限公司 The method and device of pushed information
CN108090211B (en) * 2017-12-29 2022-09-09 百度在线网络技术（北京）有限公司 Hot news pushing method and device
CN108256018A (en) * 2018-01-09 2018-07-06 北京奇艺世纪科技有限公司 A kind of item recommendation method and device
CN109145221B (en) * 2018-09-10 2021-05-28 北京一点网聚科技有限公司 Content recommendation method and device, electronic equipment and readable storage medium
CN109389168B (en) * 2018-09-29 2021-07-13 国信优易数据股份有限公司 Project recommendation model training method, project recommendation method and device
CN110969222A (en) * 2018-09-29 2020-04-07 西门子股份公司 Information providing method and system
CN110008974A (en) * 2018-11-23 2019-07-12 阿里巴巴集团控股有限公司 Behavioral data prediction technique, device, electronic equipment and computer storage medium
CN109919728A (en) * 2019-03-12 2019-06-21 贵州大学 A kind of Method of Commodity Recommendation in difference quotient city
TWI715079B (en) * 2019-06-26 2021-01-01 台達電子工業股份有限公司 Network learning system and method thereof
CN112148111A (en) 2019-06-26 2020-12-29 台达电子工业股份有限公司 Network teaching system and network teaching method thereof
CN110837813B (en) * 2019-11-14 2023-01-03 爱驰汽车有限公司 Environment equipment control method and device, electronic equipment and storage medium
CN114581165A (en) * 2020-11-30 2022-06-03 中移(苏州)软件技术有限公司 Product recommendation method, device, computer storage medium and system
CN112765467A (en) * 2021-01-19 2021-05-07 北京嘀嘀无限科技发展有限公司 Service recommendation method and device, electronic equipment and storage medium

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1756336A (en) * 2004-09-29 2006-04-05 松下电器产业株式会社 Television channel commending system and commending method
CN101431485A (en) * 2008-12-31 2009-05-13 深圳市迅雷网络技术有限公司 Method and system for automatically recommending internet information
CN102789462A (en) * 2011-05-18 2012-11-21 阿里巴巴集团控股有限公司 Project recommendation method and system

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR20130035064A (en) * 2011-09-29 2013-04-08 삼성전자주식회사 Method and system for providing personalization service based on personal tendency

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1756336A (en) * 2004-09-29 2006-04-05 松下电器产业株式会社 Television channel commending system and commending method
CN101431485A (en) * 2008-12-31 2009-05-13 深圳市迅雷网络技术有限公司 Method and system for automatically recommending internet information
CN102789462A (en) * 2011-05-18 2012-11-21 阿里巴巴集团控股有限公司 Project recommendation method and system

Also Published As

Publication number Publication date
CN103823908A (en) 2014-05-28

Similar Documents

Publication Publication Date Title
CN103823908B (en) Content recommendation method and server based on user preference
CN103678518B (en) Method and device for adjusting recommendation lists
US10409821B2 (en) Search result ranking using machine learning
CN111461841B (en) Article recommendation method, device, server and storage medium
CN110532479A (en) A kind of information recommendation method, device and equipment
CN106294830A (en) The recommendation method and device of multimedia resource
Xiang et al. Time-dependent models in collaborative filtering based recommender system
CN108205768A (en) Database building method and data recommendation method and device, equipment and storage medium
CN109493113A (en) A kind of providing method and device of favor information
CN106168980A (en) Multimedia resource recommends sort method and device
KR20140056731A (en) Purchase recommendation service system and method
CN107895213A (en) Forecasting Methodology, device and the electronic equipment of spending limit
CN106909688B (en) Method and device for recommending search terms based on input search terms
CN111768239A (en) Property recommendation method, device, system, server and storage medium
CN108595492A (en) Method for pushing and device, storage medium, the electronic device of content
CN110322281A (en) The method for digging and device of similar users
CN107103028A (en) A kind of information processing method and device
JP2018084928A (en) Selection device, selection method and selection program
JP5365948B1 (en) Content sales system and method
US20130332386A1 (en) Content management
CN106156351A (en) Multimedia resource recommendation information generates method and device
CN110362751A (en) Business recommended method, apparatus, computer equipment and storage medium
CN109102177A (en) Processing method, device and the equipment of cloud shelf
CN110020108A (en) Network resource recommended method, device, computer equipment and storage medium
CN106897412A (en) A kind of method and apparatus for recommending associated application based on intended application

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant
CF01 Termination of patent right due to non-payment of annual fee
CF01 Termination of patent right due to non-payment of annual fee

Granted publication date: 20170714

Termination date: 20210321